Data leaders and professionals are engaged in a race to prepare their organizations for robust artificial intelligence (AI) and machine learning initiatives. Notably, they are aware of and well-tuned in the roles that GenAI and large language models can serve in the AI-friendly architectures on which they are focused.
These are the findings of a March 2025 survey and analysis of 259 qualified enterprise data professionals conducted by Unisphere Research, the research arm of Information Today, Inc. The survey was done in partnership with Radiant Advisors and underwritten by Denodo, Onehouse, and Quest Software. The survey shows that emerging AI technologies are becoming an integral part of modern data architecture.
Currently, the survey finds 39.0% of enterprises are actively participating in GenAI and LLM development, with another 33.6% researching implementation strategies.
These are not foreign concepts, either—with 22.8% saying they have a detailed comprehension of the technology and another 40.9% expressing conceptual clarity about GenAI capabilities.
The business benefits are not lost on data leaders either. More than one-third (36.7%) view GenAI as “valuable for business automation and efficiency.” Only 10.8% voice concerns regarding hallucinations and security risks. This showcases an acceleration of knowledge that has transformed how data leaders evaluate all architectural components.
This strong market confidence directly correlates with realistic implementation planning. Organizations demonstrate a mature awareness of the complexities of GenAI, with only 7.7% being unfamiliar with implementation challenges—a remarkably low percentage that indicates widespread practical knowledge.
The top obstacles reflect a sophisticated understanding rather than fundamental uncertainty: LLM accuracy and hallucination risks concern 36.7% of organizations, implementation costs affect 35.5%, and enterprise data integration challenges impact 35.1%.
There are also high-level, AI-focused strategic obstacles, starting with the challenge of managing the complexity of distributed datasets and catalogs, which affects 29.7% of the organizations surveyed.
Avoiding data lock-in to specific platforms was a concern for 27%. Most tellingly, the lack of clearly defined business use cases and benefits challenges 25.9% of organizations.
Confusion about data architecture approaches impacts 20.1%. Several data platform types also emerged as preferred choices in this new, AI-enabled data architecture stack. Data lakehouses are recognized as important by 42.5% of data leaders, owing to their potential roles as flexible, open-format storage foundations that AI applications require for diverse data types and formats.
Another form of flexible architecture, data fabric, is the preferred approach of 29.3% by providing a unified data access layer across disparate sources, which are necessary for comprehensive AI training and inference. In addition, semantic layer search is seen as important by 20.1%, sought for its critical role in Retrieval Augmented Generation systems and providing the contextual metadata that enables LLMs to comprehend data relationships and meanings.
Data sources are also evolving. While traditional operational databases (i.e., Oracle, SQL, Postgres, MySQL) still dominate, as cited by 54.8%, there is a notable emergence of data types that enable AI to dominate organizational expectations.
Real-time streaming data and cloud SaaS platforms, not measured in the previous survey in 2023, came in at 49.4% and 44.4%, respectively.
Interestingly, document/semi-structured data expectations dropped from 55% to 39%. The data portfolio has become fundamentally more complex and AI-oriented, with newly measured categories such as external API data feeds (38.6%) and social data (28.2%) indicating that organizations are building architectures capable of integrating diverse, dynamic sources for comprehensive AI capabilities.